[Ocfs2-users] OCFS2 1.4 *BETA* Announcement
Sunil Mushran
Sunil.Mushran at oracle.com
Fri Mar 21 14:09:38 PDT 2008
All,
We are pleased to announce the release of OCFS2 1.4 *BETA*. We are currently
targeting production with the SLES10 SP2 and RHEL5 U2 releases, both of
which
are currently in BETA. OCFS2 1.4 *BETA* has been shipping with SLES10 SP2
BETA for sometime now. This announcement provides packages for the same for
the RHEL5 U2 BETA release.
While this is a BETA release and it is not advisable to deploy it in a
production environment, we would like to encourage you to use it in your
test
cluster and provide us with constructive feedback.
Before deploying, all users are encouraged to read the README in its
entirety,
especially the sections on the compatibilities and the new filesystem
defaults.
The packages for OCFS2 1.3.9-0.1 *BETA* for RHEL5 U2 BETA are available
here:
http://oss.oracle.com/projects/ocfs2/files/RedHat/RHEL5_BETA/
The packages for OCFS2-TOOLS 1.3.9-0.1 *BETA* for RHEL5 are available here:
http://oss.oracle.com/projects/ocfs2-tools/files/RedHat/RHEL5/i386/beta/
http://oss.oracle.com/projects/ocfs2-tools/files/RedHat/RHEL5/x86_64/beta/
As always, we look forward to hearing from you on the
ocfs2-users at oss.oracle.com
mailing list.
The OCFS2 Team
OCFS2: http://oss.oracle.com/projects/ocfs2
TOOLS: http://oss.oracle.com/projects/ocfs2-tools
README
============================================================================
WHAT'S NEW
This release makes available the features that have been steadily
accumulating
in the mainline Linux kernel tree over the past 18 months or so. The list of
features added since the 1.2 release is as follows:
* File Attribute Support
* Directory Readahead
* Performance Enhancement - stat(2)
* Performance Enhancement - unlink(2)
* Splice IO (support to be enabled by release)
* Atime/Mtime Updates
* Sparse File Support (ondisk change)
* Unwritten Extents/Punch Holes (ondisk change, support via ioctl only)
* Shared Writeable mmap(2)
* Data in Inode (ondisk change, tools support expected by release)
* Online Filesystem Resize
* Clustered flock(2)
* Ordered Journal Mode
For the full description, please refer to the following link:
http://oss.oracle.com/projects/ocfs2/dist/documentation/ocfs2-new-features.html#FEATURES
FILESYSTEM COMPATIBILITY
OCFS2 1.2 is fully ondisk compatible with OCFS2 1.4. Users installing this
release will be able to mount their existing volumes as-is.
However, one will not be able to do a rolling upgrade to 1.4 as the network
protocol has changed.
Also, users mounting existing volumes will not have access to all the new
features. Features entailing ondisk changes, like Sparse Files, Unwritten
Extents and Data-in-Inode, will need to be explicitly enabled using
tunefs.ocfs2.
Once enabled, the same volumes can still be mounted using the older 1.2
software,
but only after the said feature is disabled.
TOOLS COMPATIBILITY
Simply put, the latest OCFS2-TOOLS always support all existing versions
of the
file system. While users looking to upgrade to 1.4 (or to the latest
Linux kernel),
must install OCFS2-TOOLS 1.4, existing 1.2 users can do so at their own
convenience.
NEW FILESYSTEM DEFAULTS
Sparse file support is activated by default for volumes formatted using
the new
mkfs.ocfs2. Users wishing to retain full compatibility with OCFS2 1.2 should
specify "--fs-feature-level=max-compat" during format.
The other change in the defaults concerns the journaling mode. While
OCFS2 1.2
supported the writeback data journaling mode, 1.4 not only adds support
for the
ordered data journaling mode, it also makes it the default. Users wishing to
keep using the writeback mode can enable it by mounting the volume using the
data=writeback mount option.
The difference between the two modes is that in the ordered mode, the
file system
flushes the file data to disk before committing the metadata changes. In
writeback,
no such write ordering is preserved.
While the writeback mode guarantees internal filesystem integrity alongwith
better overall throughput than the ordered mode, there exists a
possibility of
stale data appearing in files after a crash and a journal recovery.
DISTRIBUTIONS
OCFS2 1.4 is a backport of the filesystem in the mainline Linux kernel tree,
2.6.25-rc6. It has been backported to work only on the 2.6.16 (SLES10)
and the
2.6.18 (RHEL5) kernels. No attempt has been made to ensure that it even
builds
on the other kernels.
Users using a different distribution are encouraged to use the filesystem
shipped with that distribution and not attempt to build OCFS2 1.4 on it.
There
is no reason to do so as 1.4 is simply a backport of the version in the
mainline
tree and has very little difference in functionality with it. Please
note that
as a rule, we apply bug fixes to all relevant kernel trees and not just the
enterprise kernels.
FEEDBACK
The one known missing piece in this release is CDSL support. Our
implementation
in the 1.2 release was not accepted by the Linux kernel community. While
we can
always merge back the support for only the 1.4 release, we are aiming to
make
1.4 release look as much as possible like mainline so as to make it
easier for
users to move between kernels/distributions.
With that in mind, we are looking for feedback concerning the use of
CDSL. As in,
are you using CDSL? If so, how many files? Which type of files? Answers
to the
same will help us decide our approach to the solution.
WHAT'S MISSING
Apart from CDSL support, we still need to add tools support for
data-in-inode,
enable splice io and update the documentation.
FUTURE PLANS
While not set in stone, the list of "major" features that we are/will be
working
on looks something as follows:
1. Framework for Integrating with Userspace Clusterstack(s)
This will allow OCFS2 to work with different userspace cluster stacks.
We are aiming to to push changes for the same in the upcoming 2.6.26
kernel.
2. CMAN Integration
In the coming months, the open source cluster stacks as distributed
by Red Hat
and SUSE, are likely to somewhat merge into a new CMAN that is
layered atop
OpenAIS. We will use the above framework to integrate OCFS2 with the
new CMAN.
Please note that we will continue to support the "classic" O2CB
cluster stack
and make the choice of cluster stack user configurable. For more
information,
please refer to this link.
http://sources.redhat.com/cluster/wiki/HomePage
3. Extended Attributes
This feature is currently in development. Once it is committed into
the mainline tree, we will we decide whether to backport it to 1.4 or
wait for SLES11 / (RH)EL6 to make it available on Enterprise
distributions.
4. POSIX Locking
While we managed to add support for clustered flock(2) (aka BSD locking)
in 2.6.25, we are still looking to add support for its POSIX cousin.
5. Online Adding of Node-Slots
This will allow users to increase the number of node-slots without
having to
umount the volume on all nodes. (Number of node slots dictates the number
of nodes that can mount the volume concurrently.)
6. Online Defragmentation
This will make the file system coalesce file data extents in order to
boost
performance.
7. Indexed Directories
This will offer better performance during lookups in directories
having more
than 10,000 files.
8. JBD2 Support
This will allow users to increase the volume sizes to 4PB. We will
take on
this task after ext4/jbd2 code base has somewhat stabilized.
==============================================================================
More information about the Ocfs2-users
mailing list